need faster mail server 1

[emmons]

my users are complaining about my slow mail server. I believe I have only 64 mb RAM. I am thinking I should max out the RAM to speed things up.

is this the right approach?

 

[bluegroper]

Yes.
More ram.
Faster processor. Faster HDD, faster network ?
Faster software? Maybe try postfix, qmail, or exim. I believe sendmail can be quite slow, but I don't use it.

 

[roderickom]

More RAM will only go so far, hour many users are accessing yous mail server? If there is alot of users you could use a faster processor or more bandwidth.

Roderick Shilling

http://www.orbitmicro.com

http://www.qubitmicro.com

 

[Morsing]

How many users do you have?? How much mail do you process per day??

A mail server doesn't need more than a 100MHz processor.
The memory is not the only culprit but especially the disks. A redesign of the disk layout can give you a 5-30 times performance boost. Also consider putting more disks in it.

What type of disks do you use?? How many?? What's your filesystem layout??

What mailserver are you running?? Sendmail??

Cheers

Henrik Morsing
Certified AIX 4.3 Systems Administration
& p690 Technical Support

 

[Norwich]

Have you tried running various monitoring tools to see where it is bottlenecking? Plain old "top" will show you if the server is hammering swap and needs more RAM.

If it's an IDE based system, rather than SCSI, have you looked at the "hdparm" command to optimise disk access? This can be quite spectacular if the disks aren't already (which is the case with vanilla Debian installs).

See:

http://linux.oreillynet.com/pub/a/linux/2000/06/29/hdparm.html

There is also the possibility that if you are using remote POP or IMAP clients that there is some sort of network name resolution delay that is slowing things down. This is certainly a possible if you are running Linux kit on an NT network with a Microsoft DNS server that hasn't been configured for proper reverse lookup zones (doesn't effect MS to MS comms in this case as they will 'cheat' and use WINS or broadcasts to reverse resolve if DNS is timing out).

 

[emmons]

Here is what I get with the top command at this time
What do you think?

9:36am up 23:47, 1 user, load average: 0.02, 0.06, 0.04
85 processes: 84 sleeping, 1 running, 0 zombie, 0 stopped
CPU states: 1.1% user, 1.7% system, 0.0% nice, 97.0% idle
Mem: 78280K av, 76628K used, 1652K free, 0K shrd, 5388K buff
Swap: 196520K av, 8744K used, 187776K free 11752K cached

 

[Norwich]

Well, it's using swap so it could with a bit more memory. Though it hasn't used much so 64MB should be OK.

Is this the top results when the server is running slow? There is hardly any processor load so a bigger processor isn't necessary if this is average load.

 

[emmons]

Norwich,

...but doesn't it use almost all the memory? Only 1652K is free in the above example.

I don't know if this is the top results when the server is running slow. I will have to try to find out when it is running slow and then look at top.

I would like to but I am quite afraid to try hdparm since the library would be in a world of hurt without this email server.

I don't know if you have seen my other desperate posts but most of my clients are using netscape client. Some of them get into email fine some of the time. Other times they can't ie it doesn't ask for a password or retrieve new messages. It's been acting up since the new watchguard firewall and since the dhcp server. It seems to start and stop working on its own. Any comments?

 

[Norwich]

Hi,

The 1652K free indicates that that amount is available at the moment. The used SWAP 8744K indicates that at some point, swapping was instigated and that amount was used. If a box generally uses swap space, then there's an argument to put in proper RAM.

Playing with hdparm isn't too bad - and provided you haven't placed the command in a startup script while experimenting a reboot will get it running again. Just make sure you do it in a scheduled maintenence period and run it in single user mode so no-one is unexpectedly on. I've always found the general:

/sbin/hdparm -X66 -d1 -u1 -m16 -c3 /dev/hda

never does any harm on recent systems.

How are the netscape clients talking to the mail server? POP3, IMAP4, direclty to the spool file?

 

[bluegroper]

Just stuff some more memory into it and see what happens. Only downside is a measly few bucks.

 

[emmons]

The netscape clients talk to the mail server via imap. They go through the firewall into the dmz. The mail server sees everybody as coming from the same ipaddress.

http works fine. imap doesn't in the middle of the day when all users are in the building. it only started acting up after we got the new watchguard firewall and dhcp server.

It's driving the library staff crazy. And me too.

I'd like to figure out why this is happening and fix it.

Thank you so much.

 

[bluegroper]

it only started acting up after we got the new watchguard firewall and dhcp server.
Do you think the firewall might be causing the problem ?
Have you tried the extra RAM yet ?

 

[Norwich]

Sounds like you are MASQing the internal network behind the firewall's internal interface (as the DMZ sees them as one IP address). Try reconfiguring so that the local network MASQs behind the external interface so the DMZ will recieve packets from the private machines proper IP addresses.

 

[emmons]

Alrighty then Norwich. So I need to read the manual for the watchguard firewall in order to figure out how to reconfigure it to MASQ the local network behind the external interface only instead of behind the internal interface.

ok. thanks.

 

[Morsing]

emmons, the amount of memory or paging space used won't tell you much by itself. You need to looks at the si an so (pi/po) columns from the vmstat output.

Could it be a name resolver problem?? Maybe the firewall blocks the DNS lookups from the server??

Cheers

Henrik Morsing
Certified AIX 4.3 Systems Administration
& p690 Technical Support

 

[emmons]

morsing,

You might be onto something.

I understand that the domain name server translates a name into the ipaddress and that we have an MX record. I don't know any more than that. Can you explain a little more about the firewall blocking the DNS lookups from the server.

Could you explain a little more please? Specifically, what would I do to fix that?

I discovered vmstat this weekend but wasn't sure how to interpret it.

thanks, emmons

 

[lgarner]

First, check /etc/resolv.conf and make sure that your domain and a couple of valid nameservers are listed. Then, run "nslookup" at the server console. That will pretty much emulate what the server does to resolve names. If it's slow responding, then maybe you need to change your /etc/resolv.conf to a better public (your ISP) nameserver. If it doesn't work at all, then you can check the firewall.

One option which might help is to run Bind on your server and set the first nameserver in /etc/resolv.conf to "127.0.0.1". Make sure that you put a couple of valid nameservers in the "forwarders" section of named.conf. Then your server will act as the DNS server. This helped my proxy servers considerably.

 

[emmons]

lgarner,

can I do these things from home using ssh or do I have to be physically at the server?

emmons

 

[emmons]

lgarner,

I don't have a named.conf file in my etc directory.
I do have resolv.conf

when I typed nslookup I got some message saying it was deprecated and to look at dig or host programs ?!?!?

emmons

 

[lgarner]

emmons,

You can use ssh for all of this. I frequently download files to my pc and scp them to my server if I'm away from it.

If Bind is installed, I think there should be a named.conf file in /etc as the standard location. I just checked my Suse 8 and RedHat 8 servers and that's how they both do it. Try installing it from your distribution, or go to

http://www.isc.org

for the latest version (9.2.2).